Emacs 22.1 fails to build under GNU/Linux on S/390x due to multilib assumption

Emacs 22.1 fails to build under GNU/Linux on S/390x due to multilib assumption

[Ulrich Mueller]

Hi,

this is a forwarded report of Gentoo bug 205555,
<http://bugs.gentoo.org/205555>, reported by Mike Frysinger:

> the emacs build system makes assumption about multilib setups based
> on the target architecture

> src/m/amdx86-64.h:#define START_FILES pre-crt0.o /usr/lib64/crt1.o /usr/lib64/crti.o
> src/m/amdx86-64.h:#define LIB_STANDARD -lgcc -lc -lgcc /usr/lib64/crtn.o
> src/m/ibms390x.h:#define START_FILES pre-crt0.o /usr/lib64/crt1.o /usr/lib64/crti.o
> src/m/ibms390x.h:#define LIB_STANDARD -lgcc -lc -lgcc /usr/lib64/crtn.o

> this probably isnt a big deal for amd64, but on s390x, /usr/lib64/
> does not exist in a non-multilib setup.  there is only /usr/lib/.

Below I include a patch that fixes it for S/390x.

Maybe one should also rethink if HAVE_X86_64_LIB64_DIR and the code
around it are the best possible solution on x86_64. Especially, there
seems to be no way to override it via --libdir.

Ulrich

--- emacs-22.1-orig/src/Makefile.in	2007-03-21 15:31:24.000000000 +0100
+++ emacs-22.1/src/Makefile.in	2008-01-13 03:53:26.000000000 +0100
@@ -29,6 +29,9 @@
 
 # Here are the things that we expect ../configure to edit.
 # We use $(srcdir) explicitly in dependencies so as not to depend on VPATH.
+prefix=@prefix@
+exec_prefix=@exec_prefix@
+libdir=@libdir@
 srcdir=@srcdir@
 VPATH=@srcdir@
 CC=@CC@
--- emacs-22.1-orig/src/m/ibms390x.h	2007-01-21 04:59:39.000000000 +0100
+++ emacs-22.1/src/m/ibms390x.h	2008-01-13 03:51:46.000000000 +0100
@@ -160,10 +160,10 @@
 #define XPNTR(a) XUINT (a)
 
 #undef START_FILES
-#define START_FILES pre-crt0.o /usr/lib64/crt1.o /usr/lib64/crti.o
+#define START_FILES pre-crt0.o $(libdir)/crt1.o $(libdir)/crti.o
 
 #undef LIB_STANDARD
-#define LIB_STANDARD -lgcc -lc -lgcc /usr/lib64/crtn.o
+#define LIB_STANDARD -lgcc -lc -lgcc $(libdir)/crtn.o
 
 /* arch-tag: 4b87653c-6add-4663-8691-7d9dc17b5519
    (do not change this comment) */

Re: Emacs 22.1 fails to build under GNU/Linux on S/390x due to multilib assumption

[Andreas Schwab]

Ulrich Mueller <ulm@gentoo.org> writes:

> --- emacs-22.1-orig/src/m/ibms390x.h	2007-01-21 04:59:39.000000000 +0100
> +++ emacs-22.1/src/m/ibms390x.h	2008-01-13 03:51:46.000000000 +0100
> @@ -160,10 +160,10 @@
>  #define XPNTR(a) XUINT (a)
>  
>  #undef START_FILES
> -#define START_FILES pre-crt0.o /usr/lib64/crt1.o /usr/lib64/crti.o
> +#define START_FILES pre-crt0.o $(libdir)/crt1.o $(libdir)/crti.o

This is wrong.  libdir is the directory for the package to store its
library files, not the system library directory.  Even if prefix is
/usr/local you still need to look at /usr/lib(64).

Andreas.

-- 
Andreas Schwab, SuSE Labs, schwab@suse.de
SuSE Linux Products GmbH, Maxfeldstraße 5, 90409 Nürnberg, Germany
PGP key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5
"And now for something completely different."

Re: Emacs 22.1 fails to build under GNU/Linux on S/390x due to multilib assumption

[Mike Frysinger]

[-- Attachment #1: Type: text/plain, Size: 1166 bytes --]

On Tuesday 15 January 2008, Andreas Schwab wrote:
> Ulrich Mueller <ulm@gentoo.org> writes:
> > --- emacs-22.1-orig/src/m/ibms390x.h	2007-01-21 04:59:39.000000000 +0100
> > +++ emacs-22.1/src/m/ibms390x.h	2008-01-13 03:51:46.000000000 +0100
> > @@ -160,10 +160,10 @@
> >  #define XPNTR(a) XUINT (a)
> >
> >  #undef START_FILES
> > -#define START_FILES pre-crt0.o /usr/lib64/crt1.o /usr/lib64/crti.o
> > +#define START_FILES pre-crt0.o $(libdir)/crt1.o $(libdir)/crti.o
>
> This is wrong.  libdir is the directory for the package to store its
> library files, not the system library directory.  Even if prefix is
> /usr/local you still need to look at /usr/lib(64).

i'd point out that hardcoding the start files is wrong on so many levels.  
this doesnt refute your statements here about the nature of $(libdir) (which 
are absolutely correct), just that the way the start files are discovered is 
completely screwed (just like the rest of the emacs build system).  i dont 
know how deep of hacks are acceptable, but other possibilities:
 - for $host_cpu == s390x, check /usr/lib64/ and then /usr/lib/
 - use gcc -print-file-name=crt1.o
-mike

[-- Attachment #2: This is a digitally signed message part. --]
[-- Type: application/pgp-signature, Size: 827 bytes --]

Re: Emacs 22.1 fails to build under GNU/Linux on S/390x due to multilib assumption

[Andreas Schwab]

Mike Frysinger <vapier@gentoo.org> writes:

>  - use gcc -print-file-name=crt1.o

Yes, this is the correct way.

Andreas.

-- 
Andreas Schwab, SuSE Labs, schwab@suse.de
SuSE Linux Products GmbH, Maxfeldstraße 5, 90409 Nürnberg, Germany
PGP key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5
"And now for something completely different."

Re: Emacs 22.1 fails to build under GNU/Linux on S/390x due to multilib assumption

[Glenn Morris]

Mike Frysinger wrote:

>  - for $host_cpu == s390x, check /usr/lib64/ and then /usr/lib/

So in other words, for Emacs 22.2 we can just do for s390x what we
already do for amd64?

Could someone confirm if this works:


*** configure.in      7 Jan 2008 08:53:18 -0000 1.444.2.6
--- configure.in      16 Jan 2008 19:12:52 -0000
***************
*** 1532,1538 ****
  ### The standard library on x86-64 GNU/Linux distributions can 
  ### be located in either /usr/lib64 or /usr/lib.
  case "${canonical}" in
!   x86_64-*-linux-gnu* )
      if test -d /usr/lib64; then
        AC_DEFINE(HAVE_X86_64_LIB64_DIR, 1,
          [Define to 1 if the file /usr/lib64 exists.])
--- 1532,1538 ----
  ### The standard library on x86-64 GNU/Linux distributions can 
  ### be located in either /usr/lib64 or /usr/lib.
  case "${canonical}" in
!   x86_64-*-linux-gnu* | s390x-*-linux-gnu* )
      if test -d /usr/lib64; then
        AC_DEFINE(HAVE_X86_64_LIB64_DIR, 1,
          [Define to 1 if the file /usr/lib64 exists.])


*** ibms390x.h 8 Jan 2008 04:04:33 -0000 1.4.2.3
--- ibms390x.h 16 Jan 2008 19:11:41 -0000
***************
*** 161,170 ****
--- 161,178 ----
  #define XPNTR(a) XUINT (a)
  
  #undef START_FILES
+ #ifdef HAVE_X86_64_LIB64_DIR
  #define START_FILES pre-crt0.o /usr/lib64/crt1.o /usr/lib64/crti.o
+ #else
+ #define START_FILES pre-crt0.o /usr/lib/crt1.o /usr/lib/crti.o
+ #endif
  
  #undef LIB_STANDARD
+ #ifdef HAVE_X86_64_LIB64_DIR
  #define LIB_STANDARD -lgcc -lc -lgcc /usr/lib64/crtn.o
+ #else
+ #define LIB_STANDARD -lgcc -lc -lgcc /usr/lib/crtn.o
+ #endif
  
  /* arch-tag: 4b87653c-6add-4663-8691-7d9dc17b5519
     (do not change this comment) */

Re: Emacs 22.1 fails to build under GNU/Linux on S/390x due to multilib assumption

[Mike Frysinger]

[-- Attachment #1: Type: text/plain, Size: 354 bytes --]

On Wednesday 16 January 2008, Glenn Morris wrote:
> Mike Frysinger wrote:
> >  - for $host_cpu == s390x, check /usr/lib64/ and then /usr/lib/
>
> So in other words, for Emacs 22.2 we can just do for s390x what we
> already do for amd64?
>
> Could someone confirm if this works:

i just tested on an s390x/Gentoo system and it works for me, thanks.
-mike

[-- Attachment #2: This is a digitally signed message part. --]
[-- Type: application/pgp-signature, Size: 827 bytes --]

Re: Emacs 22.1 fails to build under GNU/Linux on S/390x due to multilib assumption

[Ulrich Mueller]

>>>>> On Wed, 16 Jan 2008, Glenn Morris wrote:

>       if test -d /usr/lib64; then
>         AC_DEFINE(HAVE_X86_64_LIB64_DIR, 1,
>           [Define to 1 if the file /usr/lib64 exists.])

So the existence of a /usr/lib64 directory implies that crt1.o is
located there?

Ulrich

Re: Emacs 22.1 fails to build under GNU/Linux on S/390x due to multilib assumption

[Mike Frysinger]

[-- Attachment #1: Type: text/plain, Size: 886 bytes --]

On Thursday 17 January 2008, Ulrich Mueller wrote:
> >>>>> On Wed, 16 Jan 2008, Glenn Morris wrote:
> >
> >       if test -d /usr/lib64; then
> >         AC_DEFINE(HAVE_X86_64_LIB64_DIR, 1,
> >           [Define to 1 if the file /usr/lib64 exists.])
>
> So the existence of a /usr/lib64 directory implies that crt1.o is
> located there?

at the moment, yes.  emacs assumes glibc and it assumes that the only time   
you have /usr/lib64/ is if you're running on a system (multilib or not) that 
is 64bit native.  it's an OK assumption given the overall limitations in the 
emacs build system and should effectively work on pretty much all distros out 
there.  if you want to talk about straying outside the norm, the current 
system does not scale at all and would require a rewrite, so trying to solve 
the s390x issue at the sametime isnt really the way to go.
-mike

[-- Attachment #2: This is a digitally signed message part. --]
[-- Type: application/pgp-signature, Size: 827 bytes --]